A Study on Predicting the Resistance Value of Indium Tin Oxide Films by Model Trees and Partial Least Squares

• Main Authors: Chih-ChengHuang, 黃誌成
• Other Authors: Tzu-Tsung Wong
• Format: Others
• Language: zh-TW
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/w89m5g

碩士 === 國立成功大學 === 工業與資訊管理學系碩士在職專班 === 105 === The resistance value of a touch panel highly depends on the thickness of the indium-tin-oxide film in its the surface. Engineers can set several manufacturing parameters, or attributes, of a sputtering machine to control the thickness of the film, such as vacuum pressure, gas flow, DC power, and transmission speed. Since the sputtering process of touch panels is continuous, their resistance values may vary due to the offset of the manufacturing parameters. The cost spent in finding and recovering abnormal products can be high. This study will therefore employ model tree and partial least square to discover the attributes that can affect the resistance values of touch panels. In model trees, the branching attributes and the attributes showing in linear regression models reveal critical manufacturing parameters. Partial least square will rank attributes based on their contribution levels in predicting resistance values of touch panels. The data collected from sputtering machines are first processed by the tools for attribute selection and normalization. Then both model tree and partial least square are applied to discover key attributes. The experimental results of these two tools for numeric prediction suggest that ‘chamber temperature’ and ‘DC power’ are the most critical manufacturing parameters for the resistance values of touch panels. Their coefficients given in linear regression models provide useful information for engineers to set proper parameters of sputtering machines.